model-based RL

World models have quietly become one of the most consequential ideas in machine learning. A system that learns how an environment behaves, rather than just mapping input to output, can predict what happens next, simulate possible outcomes and plan actions before committing to any of them.